And as long as we're complaining about compute-heavy problems, 2016 day 5 is the worst; it took 3 hours and 11 minutes to complete in m4 on my hardware (computing md5 in python involves calling into a library which in turn probably uses pre-compiled C code rather than native python for the core computations, but computing it natively in m4 is just painfully slow). Source: about 3 years ago

So I coded up something that I did not see mentioned anywhere in the megathread. Instead of storing just the recipe number in each array location, with an array entry for every recipe, I stored a tuple of recipe number and next array position. For a0-a22, the next number is pre-computed. But for a23 onwards, the next array position is exactly one larger - that is, the rest of my array stores ONLY the recipes... Source: about 3 years ago

So my next round did what many of the megathread solutions did: for each pair of input lines, determine if they are similar. This cuts the complexity from exponential O(2^n) to quadratic O(n^2), and runtime was down to 1.3s (a 300x speedup over regex abuse). Note that due to the use of substr, I risk inadvertent macro expansion if slicing a lower-case line forms a macro name such as 'nl', so I had to further... Source: about 3 years ago

Wow, that really made a difference! I got another ~250x speedup (best-case run from ~18.5s to 71ms) with this tweak. Instead of performing a full matrix multiply (O(1) work per iteration regardless of active rows, and O(log n) iterations), I take advantage of the sparseness of the initial few iterations. That is, I start out with only one row active; (however, by iteration 50 all 92 rows are active), as well as... Source: over 3 years ago

Or course, only after I had written my entire solution did I notice in the megathread that avoiding division when computing math modulo M31 (0x7fffffff) is already a known optimization. But I thought it was pretty cool that I was able to derive that optimization from something more general, and that my solution works without any 64-bit math or explicit division operators. Source: about 3 years ago

And much of the code written by scientists using government grants has to be open source by law and there's a site where you can view it all. Source: over 1 year ago

There's the worker bee model. You find a company dedicated to FOSS like Google, RedHat, Intel, etc., join them, and work your way into a position where you're maintaining something open source but salaried through your employer. There are smaller companies where this may apply, too, although the big ones are of course those that may jump to mind. You'll also find some open source opportunities within the... Source: almost 3 years ago

I do like the trend of new government projects open sourcing their systems, like https://code.gov. Source: about 3 years ago